GAN Applications

I’ve come across GANs and the most popular use case I’ve seen is artificial face generation. I know there are more applications and so I am wondering what those would be today.

  1. What are some applications GANs are presently being explored?
  2. In the GAN architecture, is it always the goal for the generator to overpower the discriminator in order to generate new artificial data that looks real?
  3. Have GANs been used to generate new data to be used in training?
  1. GANs can be used for more than generating faces. It can be used for image/music/video synthesis, style transfer, in-painting, uncropping and colorization.
  2. The goal is to reach an equilibrium and neither generator nor discriminator should over power the other. In general the discriminator (a binary classifier) has an easier job though.
  3. Using GANs for data augmentation depends on how good your GAN model is, and the task at hand. In theory it should work although sometimes the synthetic data may or may not help.